You still need to walk the tree, so you don't know which pages you need immediately- but since the root pages for the indexes should be well-distributed in the swarm, it shouldn't take very long to figure out which leaves you need.
What bothers me is that you've rounded the time down in your calculation despite ignoring throughput factors like packet overhead, piece message metadata, other protocol messages and processing time. I'd be interested to see some real time measurements. If you decide to do it (maybe for a research paper?), consider using kibis explicitly (and mind capitalization - b vs B).